Ridge-furrow Detection in Glycine Max Farm Using Deep Learning

We propose ridge-furrow detection for row-crop farms using DeepLab, which can be trained by RGB color information. Detected images are post-processed as intermediate information that facilitates furrow detection and ridge localization. Using the proposed model, the resultant prediction accuracy is about 84.65%. We deploy the transferred, learned model in a mobile base that navigates along detected furrows of a glycine max farm video. We use row-crop detection information to estimate navigation trajectories and generate line velocity and angular velocity parameters for the mobile base. During navigation of the mobile base, the derived information can be fused for use in extended agricultural tasks such as weeding and other farm labor.